A Recoverable Summer-Temperature Signature in Relative Wing Size of the Damselfly Lestes sponsa Out-of-population inverse prediction and spatial-block validation of an environmental morphological signal

Abstract

Environmental effects on insect wing morphology are well established, and geographic variation in Lestes sponsa wing size and shape has already been described [1-9]. Here we ask a distinct inverse question: can summer temperature be recovered from wing morphology of a population excluded from model fitting? We reanalyzed the open 17-population L. sponsa geometric-morphometric dataset [1] and linked population coordinates to CHELSA 1981-2010 climatologies [22]. The source-defined wing ratio (wing centroid size/body size) was evaluated as a low-dimensional relative-wing-size channel, alongside Procrustes landmark geometry. Leave-one-population-out (LOPO) prediction recovered August temperature from wing ratio in both forewings (r=0.780, MAE=1.61 °C) and hindwings (r=0.774, MAE=1.63 °C). Full 14-landmark geometry generalized poorly, whereas exploratory localization concentrated fitted temperature-associated deformation near LM9; LM9 alone was not predictive. To address spatial dependence [23,24], we applied geographic-block cross-validation. Under the strict five contiguous-latitude-block design, performance declined (forewing r=0.428, hindwing r=0.419), but permutation tests of the full blocked workflow remained better than random (P=0.0125 and 0.0195). We therefore identify a recoverable summer-temperature signature in a sparse, low-dimensional component of damselfly wing morphology dominated by relative wing size. The analysis does not establish that temperature directly causes the observed morphology, nor that the wing is a literal biological thermometer. Its novelty lies in treating wing morphology as an inverse environmental prediction problem and testing transfer to unseen populations. Keywords: Odonata; Lestes sponsa; relative wing size; geometric morphometrics; temperature; inverse prediction; environmental signature; spatial cross-validation; phenotypic plasticity